Optimal Clustering Framework for Hyperspectral Band Selection (1904.13036v1)

Abstract: Band selection, by choosing a set of representative bands in hyperspectral image (HSI), is an effective method to reduce the redundant information without compromising the original contents. Recently, various unsupervised band selection methods have been proposed, but most of them are based on approximation algorithms which can only obtain suboptimal solutions toward a specific objective function. This paper focuses on clustering-based band selection, and proposes a new framework to solve the above dilemma, claiming the following contributions: 1) An optimal clustering framework (OCF), which can obtain the optimal clustering result for a particular form of objective function under a reasonable constraint. 2) A rank on clusters strategy (RCS), which provides an effective criterion to select bands on existing clustering structure. 3) An automatic method to determine the number of the required bands, which can better evaluate the distinctive information produced by certain number of bands. In experiments, the proposed algorithm is compared to some state-of-the-art competitors. According to the experimental results, the proposed algorithm is robust and significantly outperform the other methods on various data sets.

• The paper introduces an Optimal Clustering Framework (OCF) that uses dynamic programming for efficient hyperspectral band selection by finding optimal clustering structures.
• Key methodological advancements include the Rank on Clusters Strategy (RCS) for selecting representative bands and an automatic variance-based method for determining the optimal number of bands.
• Experimental results on multiple datasets demonstrate the proposed method's superior overall accuracy and computational efficiency compared to existing approaches in hyperspectral data analysis.

Optimal Clustering Framework for Hyperspectral Band Selection: An Overview

The paper "Optimal Clustering Framework for Hyperspectral Band Selection" presents a novel approach to address the issue of band selection in hyperspectral images (HSIs). This research introduces an optimal clustering framework (OCF) designed to efficiently achieve optimal band clustering results, which greatly aids in the band selection process. The authors propose a three-pronged contribution: the development of OCF, a rank on clusters strategy (RCS) for effective band selection, and an automatic band number determination method.

Methodological Advancements

1. Optimal Clustering Framework (OCF): At the core of the paper, OCF targets the identification of optimal clustering structures by leveraging a dynamic programming approach. This framework is defined under the assumption that bands with contiguous indices, due to their correlation in wavelength, can effectively reduce the search space for clustering. The framework is adaptable to various forms of objective functions, ensuring its flexibility in application.
2. Rank on Clusters Strategy (RCS): RCS is introduced as a novel method to select representative bands within the derived clustering structure. By ranking the bands and selecting based on these rankings from each cluster, RCS ensures that the selected bands maintain low correlation and enhanced discrimination capabilities relative to the hyperspectral data.
3. Automatic Band Number Determination: The paper advocates for a method to determine the number of bands required, which reduces potential redundancy. This is achieved using a variance-based criterion, providing a more informative depiction of the distinctive information that can be derived from specific band numbers.

Experimental Validation

Experimental results carried out on multiple hyperspectral data sets demonstrate the robustness and effectiveness of the proposed method in comparison with several existing approaches. The OCF exhibits superior performance across different classifiers, maintaining consistent and high overall accuracy (OA) values. The empirical findings suggest that the proposed algorithms are computationally efficient while ensuring performance improvements over comparable state-of-the-art methods.

Implications and Future Directions

This research has significant implications in hyperspectral imaging analysis. By enabling more efficient band selection, the framework potentially enhances various applications such as remote sensing, biological analysis, and medical imaging. The improvement in computational efficiency and classification accuracy can lead to more accurate interpretations and analyses of hyperspectral data.

Moving forward, potential areas for development include exploring more complex similarity matrices that capture the intrinsic geometry of HSIs and the development of enhanced objective functions for the OCF. Such future work could further refine the balance between efficiency and accuracy in hyperspectral band selection.

In conclusion, the proposed framework has set a solid foundation for advancing hyperspectral band selection methodologies, offering a promising direction for both theoretical exploration and practical application.